sbondaryev

Thanks for sharing!

An interactive guide to how attention, embeddings, and positional encoding work together to predict the next move.

Inspired by 3Blue1Brown's Wordle video I applied maximum entropy to Mastermind.

The solver averages 4.42 guesses across all 1296 possible codes.

Includes an interactive game where you can see the optimal next move.

That's fair. It's just a suggestion - it's been useful for me in early stages of UI prototyping.

Nice project!

This pairs nicely with ASCII-Driven Development - for iterating and modifying layouts with AI.

https://medium.com/@calufa/ascii-driven-development-850f6666...

https://sbondaryev.dev/

An attempt to build intuition with interactive articles and experimentation, inspired by explorabl.es

Hi HN, I built a chess position recognition system that runs entirely in the browser.

The project walks through the full pipeline: manually collecting data from book scans, comparing Autoencoders vs. SimCLR for representation learning, and deploying with ONNX Runtime and Pyodide to avoid server costs.

The included demo recognizes board diagrams and generates Lichess links for analysis.

Nice project! One small suggestion, adding a search or category filter would help simplify navigation given the number of calculators available.

Nice touch - seeing the Windows 95 IE favicon took me back for a while.

Seems like reading the code is now the real work. AI writes PRs instantly but reviewing them still takes time. Everything flipped. Expect more projects to follow - maintainers can just use ai themselves without needing external contributions.

https://sbondaryev.dev/

An attempt to build intuition with interactive articles and experimentation, inspired by explorabl.es

Nice visualization of the residual connections. Is the animated svg manually created or programmatically generated? What tools did you use?

It's a website https://sbondaryev.dev/

I'm working on interactive explorations of algorithms and machine learning. Small, visual, hands-on demos that help build intuition by letting people tweak code/parameters and see how things behave in real time.

This looks promising - I've always found the schema duplication annoying with Zod/Yup. Would be great to see benchmarks vs Typia/ArkType at https://moltar.github.io/typescript-runtime-type-benchmarks/

Great idea, thanks for sharing! Tried your prompts with ChatGPT and Claude than iterated on it. The ASCII doesn't render perfectly in the web interface but looks good when copy/pasted into a text editor. Key benefit: I used to iterate on layout by generating HTML+Tailwind directly, which burns tokens fast. This ASCII approach lets you nail the structure first without the token cost. Much better for free tier usage. Appreciate the writeup!

It really depends on your level - this kind of graphics work is usually learned over many years.

Some sources

Math & motion: – The Nature of Code https://natureofcode.com/ – Coding Math https://www.youtube.com/user/codingmath

Shaders / math-based rendering: – The Book of Shaders https://thebookofshaders.com/ – Inigo Quilez https://iquilezles.org/

Interactive explanations: – Red Blob Games https://www.redblobgames.com/

You can also find some insight into his work process here: https://x.com/BCiechanowski/status/1387827101294686210?s=20

Most people doing this level of work built their own tools over time, learned a bit here and there, and kept refining things throughout their career.

Nice project! I'm not very strong at note recognition, so I see the “Wrong note” message quite often (which is totally fine), but the frequent blinking feels a bit distracting. You might consider making the feedback a little less visually intense.

I like this book. The probability section is great, especially how they handle the Monty Hall paradox. They use "four step method" that breaks it down perfectly - way clearer than the explanations you get in movies like 21 or numb3rs.

I skimmed the source code (base.js, light.js) to see what he was using. It appears to be entirely custom, with no graphics libraries like Three.js. He even implements his own low-level math functions from scratch. It's impressive to see that kind of discipline.

The examples are great. It would be really nice to have a sandbox with the full training code (e.g. MNIST) to play with.

Really nice work - love the aesthetic. It'd be great to have this as a JS library with support for arbitrary audio inputs.

Nice idea - unfortunately I didn't manage to get it running on my MacBook Pro (Late 2013).

I'm on Big Sur 11.7.10, which I know is pretty old at this point. The app launches, but it only shows an empty window.

I was also a bit concerned about privacy while running it :)

Nice project — it makes long HN threads much easier to revisit.

Interesting to see the edge cases, like "Tell HN: Happy New Year": https://hn-discussions.top/happy-new-year-2025/

Basically everyone just saying "Happy New Year" so it's fun to see how the LLM tries to interpret that. For example, my comment was summarized as: "gratitude – celebratory – looking forward: celebratory greeting for the new year"

First programs in basic, then I made a calculator in C using Kernighan & Ritchie book. Did some projects in Delphi/Maple at university, after graduation I started working in software companies and learned a lot from more experienced devs.

Pretty cool how the model can improve the SVG step by step. I didn’t use api-key but checked the gallery. Good idea to show prices and different LLMs on the same prompt (kind of benchmarking). One thing — I think the "API not shared" note isn’t very visible, maybe make it clearer

yeah, i learned this the hard way. I tried producthunt, indie hackers, linkedIn, twitter even instagram thinking distribution was the issue.

Turns out it was mostly audience and presentation. Math/educational content just doesn't travel well on those channels unless it’s tailored very carefully. Being right wasnt enough if the delivery didn’t match how people consume it.

Law 20: A good design with a bad presentation is doomed immediately

I definitely struggle with this. I run a math education site and I usually focus heavily on technical accuracy but underestimate the presentation.

Hard lesson that being "right" isn't enough if the delivery is clunky.

This reminded me of retina screenshots on mac — selecting a 100×100 area can produce a 200×200 file. Different cause but same idea - the stored pixels don’t always match what you see on screen.

Happy New Year from Switzerland!